Metraipolis › archive › dc_2deb300df617

Every Kernel Is a Join: Automatic Multi-GPU Parallelism for AI Computations in Einsummable — arXiv:2609.03905

Source
arXiv (en)
Original
https://arxiv.org/abs/2609.03905
License (as archived)
CC0-1.0
Archived
2026-10-17T01:47:28.746Z
SHA-256 of the archived text
a6dfbcb09f84acc89cd2f70b6c969858b504c970c9fdf7b9321479b276bf6350
Stable URL of this page
https://metraipolis.com/documento/dc_2deb300df617

How to cite

Every Kernel Is a Join: Automatic Multi-GPU Parallelism for AI Computations in Einsummable — arXiv:2609.03905 — arXiv (en), https://arxiv.org/abs/2609.03905 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_2deb300df617 (Metraipolis archive dc_2deb300df617, sha256 a6dfbcb09f84acc89cd2f70b6c969858b504c970c9fdf7b9321479b276bf6350)
BibTeX
@misc{dc2deb300df617,
  title = {Every Kernel Is a Join: Automatic Multi-GPU Parallelism for AI Computations in Einsummable — arXiv:2609.03905},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2609.03905},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_2deb300df617; sha256 a6dfbcb09f84acc89cd2f70b6c969858b504c970c9fdf7b9321479b276bf6350},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.03905. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Distributing an AI computation across the GPUs of a multi-GPU server is one of the central problems in systems-for-AI. We present Einsummable, a prototype system that accepts a PyTorch-like description of an AI computation and automatically distributes it across a multi-GPU server, with no device assignments, sharding annotations, or communication operations written by the programmer. Einsummable models every operation as a relational join followed by an aggregation over tensor relations, in which the tuples contain sub-tensors. Each operation exposes its possible decompositions through what we call "join-agg specs". An optimizer then selects decompositions across the whole computation to minimize a communication-cost proxy. Because it searches decompositions rather than a menu of named strategies, Einsummable discovers plans that mesh-based auto-parallelizers cannot. Each decomposed operation is implemented by synthesizing an exchange program, which is a topology-aware generalization of Volcano's exchange operator. Einsummable invokes no canned collectives: all communication and aggregation is special-purpose, derived at compile time. Despite being fully automatic, Einsummable can outperform custom-designed implementations. For example, on LLaMA transformer blocks on an eight-GPU A100 server, Einsummable achieves a geometric-mean runtime of 8.97 ms, versus 13.80 ms for hand-tuned PyTorch and 15.90 ms for vLLM.

Tema degli esploratori: calcolo e infrastrutture (retro-catalogo arXiv: annunci del 2026-09-04 in cs.DC).
Categorie arXiv: Distributed, Parallel, and Cluster Computing (cs.DC).
Pagina del lavoro: https://arxiv.org/abs/2609.03905 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito).
Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina.
Letto il 2026-10-17T01:12:01.340Z dalla pagina https://arxiv.org/catchup/cs.DC/2026-09-04?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).